Two approaches to design of forward adaptive piecewise uniform scalar quantizers

Abstract

In this paper, two forward adaptive piecewise uniform scalar quantizers are proposed for high-quality quantization of speech signals modeled by the Laplacian probability density function. In designing both forward adaptive piecewise uniform scalar quantizers an equidistant support region partition is assumed and a distribution of the number of reproduction levels per segments is optimized. The proposed models differ in the approach of determining the reproduction levels. In particular, one model defines the reproduction levels as the cell centroids and the other one as the cell midpoints. We show that, in the high-resolution case, the proposed quantizers provide approximately the same performance being close to the one of the forward adaptive nonlinear scalar compandor with equal number of quantization levels.